Non Probability Sampling
Chapter -Eight
Syllabus topic 5, "Research Methods"
Pages 392 to 395 of 543
In one line
In non probability sampling the chance of any unit being selected is unknown, so the results describe the people who were studied and nobody else, and that is a limitation to be stated rather than a reason not to use it.
In the wording a student can write in an exam: non probability sampling comprises those methods in which units are selected by the researcher's judgment, convenience or accessibility rather than by chance, so that the probability of inclusion is unknown and the findings cannot be generalised to the population or given a margin of error; its principal forms are convenience sampling, purposive or judgment sampling, quota sampling and snowball sampling, and it is legitimate where a probability sample is impossible, where the study is exploratory or qualitative, or where the population cannot be enumerated.
Convenience sampling
What it is. Taking whoever is available: the people at the court gate, the students in one class, the respondents who agreed.
Its merits. It is fast, cheap and sometimes the only thing possible.
Its demerits. The sample is systematically unlike the population, because availability is not random: the people at the court gate at eleven o'clock are the ones whose matter was listed and who could come.
And it is the commonest error in student legal research. Chapter 690 named it: interviewing whoever was willing and then writing as though the result described the district. The method is not the error; the generalisation is.
When it is legitimate. In an exploratory stage, to find out what should be asked, chapter 820; in a pre-test, chapter 690; and where the study says plainly what it is.
Purposive or judgment sampling
What it is. Deliberately choosing units because of what they are: the three colleges known to run real clinics; the magistrates with the longest remand lists; the district with the highest disposal rate.
Its merits. It gets exactly the units the question needs, which random selection may miss entirely. For studying an unusual phenomenon it is the only sensible method.
Its demerits. The result depends on the researcher's judgment about what is relevant, and a researcher expecting a conclusion will choose units that produce it.
When it is legitimate. Case studies, chapter 980; studies of extremes, of best practice, or of a defined institutional set; and any study whose claim is about the units chosen rather than about a population.
Quota sampling
What it is. Deciding in advance how many of each category are wanted, then filling those quotas by any means: forty women and forty men, or thirty from each of three settlements.
Its merits. It guarantees that categories appear, which is stratification's advantage; it needs no frame; and it is quick.
Non Probability Sampling
Its demerits. Within each quota the selection is by convenience, so the same bias operates inside every cell. It resembles stratified sampling and is not it, and confusing the two is a common examination error.
The distinction to be able to state. Stratified sampling selects randomly within each stratum from a frame; quota sampling fills each cell by whoever is available. One permits inference; the other does not.
Snowball sampling
What it is. Initial respondents are asked to name others like them, and the sample grows through their referrals.
Its merits, and this is the method's genuine claim. Some populations cannot be enumerated and cannot be reached any other way: people who tried to obtain legal aid and gave up; undertrials released and dispersed; people who used a tout; migrants without documents. There is no frame for any of them.
And in legal research these are precisely the populations that matter, because they are the people the system failed, and they are invisible to any method that starts from a list.
Its demerits. The sample follows social networks, so it over-represents the connected and misses the isolated. The first respondents shape everything that follows. And confidentiality is delicate, since respondents know one another.
When it is legitimate. When the population is hidden or unlistable, and when the study says so.
The rule that governs all four
State the method, and state what follows from it.
A study using a non probability sample may say: this is what the eighty-four people I spoke to said; here is how they were selected; here is what that selection is likely to have missed; and here is the hypothesis it suggests for testing on a probability sample.
It may not say: this is what people in the district think.
And it may not report a margin of error, because there is no basis for one.
A study that states its method honestly is publishable; a study that conceals it is not, and the difference costs one paragraph, chapter 1280.
Distinctions
| Probability | Non probability | |
|---|---|---|
| Chance of inclusion | Known and non-zero | Unknown |
| Generalisable | Yes, within limits | No |
| Sampling error estimable | Yes | No |
| Needs a frame | Usually | No |
| Cost | Higher | Lower |
| Proper use | Establishing magnitude | Exploring, reaching hidden populations, studying chosen units |
And the pairing that works. Non probability first to find out what to ask and whom to ask about, chapter 820; probability afterwards to establish how common it is.
A worked example
A study of people who tried to obtain legal aid and did not get it.
Why no probability method is available. There is no list. The District Authority records those it assisted, and a person who was refused, or who left the office before applying, or who never reached it, appears nowhere. The population is defined by its absence from every frame, which is the exact situation snowball sampling exists for.
Non Probability Sampling
How it would be done. Begin with three people identified through a legal services clinic, a settlement level worker and a para-legal volunteer; ask each, at the end, whether they know anybody else who tried and did not get help; follow the referrals until new names stop producing new kinds of account.
What it can claim. That these twenty-two people had these experiences; that the reasons they gave fall into these five kinds; and that two of the five were not anticipated.
What it cannot claim. Any proportion at all.
And what it is for. The five kinds of reason become the response options of a schedule, chapter 970, administered to a probability sample of clinic users, chapter 1010, which can then say how common each is.
That sequence is the correct use of a non probability method, and it is also the only way this particular question can be studied at all.
Quick revision
Non probability sampling: the chance of inclusion is unknown, so results cannot be generalised and no margin of error may be reported.
Convenience: whoever is available; fast and cheap; systematically unlike the population; legitimate for exploration and pre-testing. The commonest error in student legal research is not using it but generalising from it.
Purposive or judgment: units chosen for what they are; gets exactly what the question needs; depends on the researcher's judgment.
Quota: categories filled to a target by any means; guarantees categories and needs no frame; is not stratified sampling, because selection within each cell is by convenience.
Snowball: respondents name others; the only way to reach hidden or unlistable populations, which in legal research are the people the system failed; over-represents the connected.
The rule: state the method and what follows from it. The pairing that works is non probability first to find out what to ask, probability afterwards to establish how common.
Test yourself
1. What single property distinguishes non probability from probability sampling, and what two consequences follow? That the probability of any unit being included is unknown. It follows that the findings cannot be generalised to the population, and that no margin of error or sampling error can be estimated.
2. Distinguish quota sampling from stratified sampling. Both ensure that categories appear in the sample. Stratified sampling selects randomly within each stratum from a frame, so inference remains possible; quota sampling fills each category by whoever is available, so the convenience bias operates inside every cell and inference is not possible.
Non Probability Sampling
3. Why is snowball sampling especially important in legal research? Because some of the populations that matter most cannot be enumerated: people who tried to obtain legal aid and gave up, released undertrials, people who used a tout, migrants without documents. They appear on no list precisely because the system failed them, so a method that grows through referrals is the only way to reach them.
4. What may a study using a convenience sample properly claim, and what may it not? It may state what the people studied said, how they were selected, what that selection is likely to have missed, and what hypothesis the results suggest for testing on a probability sample. It may not claim to describe the population, and it may not report a margin of error.
The rest of this subject
These notes are cut from the University's printed syllabus. Open the syllabus itself, or the past papers, for the same subject.